Catastrophe Modelling and Climate Change
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Key Contacts Trevor Maynard Head of Exposure Management & Reinsurance Telephone: +44 (0)20 7327 6141 trevor.maynard@lloyds.com Nick Beecroft Manager, Emerging Risks & Research Telephone: +44 (0)20 7327 5605 nick.beecroft@lloyds.com Sandra Gonzalez Executive, Emerging Risks & Research Telephone: +44 (0)20 7327 6921 sandra.gonzalez@lloyds.com Lauren Restell Exposure Management Executive Telephone: +44 (0)20 7327 6496 lauren.restell@lloyds.com Contributing authors: Professor Ralf Toumi is Professor of Atmospheric Physics at the Department of Physics, Imperial College London. He is also a director of OASIS LMF Ltd which promotes open access for catastrophe modelling. Lauren Restell has worked in catastrophe modelling in the London market since 2006. Prior to joining Lloyd’s Exposure Management, she held positions with Aon Benfield and Travelers following completion of an MSc in Climate Change at the University of East Anglia. Case studies: The case studies were provided by Professor James Elsner (Climatek), Madeleine-Sophie Déroche (Climate – Knowledge Innovation Centre), Dr Ioana Dima and Shane Latchman (AIR), Professor Rob Lamb, Richard Wylde and Jessica Skeggs (JBA), Iain Willis (EQECAT) and Dr Paul Wilson (RMS). Acknowledgements: Lloyd’s would also like to thank Matthew Foote (Mitsui Sumitomo), Steven Luck (W. R. Berkley Syndicate) and Luke Knowles (Catlin) for their input to the report. Disclaimer This report has been produced by Lloyd's for general information purposes only. While care has been taken in gathering the data and preparing the report, Lloyd's does not make any representations or warranties as to its accuracy or completeness and expressly excludes to the maximum extent permitted by law all those that might otherwise be implied. Lloyd's accepts no responsibility or liability for any loss or damage of any nature occasioned to any person as a result of acting or refraining from acting as a result of, or in reliance on, any statement, fact, figure or expression of opinion or belief contained in this report. This report does not constitute advice of any kind. © Lloyd’s 2014 All rights reserved
Catastrophe Modelling and Climate Change Contents 1 EXECUTIVE SUMMARY 4 2 INTRODUCTION 5 3 THE SCIENCE OF CLIMATE CHANGE 6 4 CATASTROPHE MODELLING 8 5 CATASTROPHE MODELS AND CLIMATE CHANGE 10 6 CONCLUSIONS AND RECOMMENDATIONS 32 7 APPENDICES 33 8 REFERENCES 36 3
Catastrophe Modelling and Climate Change 1 Executive Summary Scientific research points conclusively to the existence of climate change driven by human activity. Nevertheless, significant uncertainty remains on the nature and extent of the changes to our climate and the specific impacts this will generate. Many of the effects will become apparent over the coming decades and anticipating them will require forward projections, not solely historical data. Changes in the climate and weather patterns have the potential to affect extreme weather events. Insurers have a key interest in understanding the impact of climate change on the frequency of extreme weather events. The frequency of heat waves has increased in Europe, Asia and Australia and more regions show an increase in the number of heavy precipitation events than a decrease. It is virtually certain that since the 1970s there has been an increase in the frequency and intensity of the strongest tropical cyclones in the North Atlantic basin. Catastrophe modelling technology is now used extensively by insurers, reinsurers, governments, capital markets and other financial entities. They are an integral part of any organisation that deals with natural catastrophe risk and are used most commonly to perform activities such as risk selection and underwriting, reserving and ratemaking, development of mitigation strategies, design of risk transfer mechanisms, exposure and aggregate management, portfolio optimisation, pricing, reinsurance decision-making and capital setting. The models help to quantify our understanding of the natural world. Climate change trends may be implicitly built into catastrophe models, given the heavy use of historical data in constructing them; however these TRENDS are not necessarily explicitly incorporated into the modelling output. Uncertainties associated with the estimation of the extent and frequency of the most extreme events means that the climate change impact can be difficult to account for in risk models. The sensitivity of hurricane losses is influenced by a number of factors related to climate change, such as sea-level rise and sea surface temperature. There is a relationship between sea surface temperatures and hurricane strength which suggests a gradual increasing trend. It is thus imperative that changes in these are modelled accurately. The approximately 20 centimetres of sea-level rise at the southern tip of Manhattan Island increased SUPERSTORM Sandy’s surge losses by 30% in New York alone. Further increases in sea- level in this region may non-linearly increase the loss potential from similar storms. Catastrophe models that dynamically model surge based on current mean sea-level already factor this increased risk into their projections. Climate models continue to project impacts on extreme weather in the coming decades. EQECAT show how future climate scenarios could see increases in the frequency of intense storms in Europe, with a possible shift in storm track towards northern latitudes. JBA notes that climate change has already increased the probability of flood events in the UK such as those which occurred in 2000, and a 1 in 5 rainfall event could be 40% larger in future. 4
Catastrophe Modelling and Climate Change 2 Introduction The insurance industry has in recent years incurred major losses as a result of extreme weather. 2011 is regarded as a record year for natural catastrophe, with insured losses costing the industry more than $127 i billion . A series of catastrophes at the end of the 1980s and the beginning of the 1990s posed a major challenge to the insurance industry. The adoption of natural catastrophe models in the 1990s helped the industry to analyse and measure risk more accurately, and use of these tools has now become the norm. Given the prevalence of catastrophe models in insurance and the rising cost of extreme weather events, the accuracy of modelled outputs is a key interest for insurers. The potential for climate change to drive changes in the severity and likelihood of extreme weather events could have implications for the accuracy of natural catastrophe models, and this report examines whether and how catastrophe models account for climate change through a series of case studies provided by a range of academic and commercial model providers. The Earth’s global climate system is warming. This conclusion is supported by a large body of evidence which is presented in the scientific literature and most comprehensively in the five Assessment Reports 1 published by the Intergovernmental Panel on Climate Change (IPCC) . Increasing greenhouse gas concentrations in the atmosphere, largely due to human activity such as combustion of fossil fuels and land use change, result in an enhancement of the planet’s natural greenhouse effect and in increased surface warming. The additionally captured energy is stored to the largest part in the oceans and, in combination with a warming of surface air temperatures, results in changes to the physical climate system. One example is the impact on the hydrological cycle in the form of changed rainfall, in changes to atmospheric circulation and weather patterns, in a reduction of global ice and snow coverage and in thermal expansion of the oceans and subsequent sea level rise. These trends challenge insurers to examine both the economic impact of climate change and the adequacy of the tools used to measure and price risks. One of the primary concerns for insurers is the potential for these changes in climate and weather patterns to affect extreme weather events. The Fourth Assessment Report of the IPCC (IPCC, 2007) highlighted the importance of our understanding of extreme events, due to their disproportionate impact on society and ecosystems when compared with gradual changes in the average climate. In 2012 the IPCC published a Special Report focusing specifically on managing the risks of extreme climate events (IPCC, 2012, from now on referred to as SREX) and the recently released draft of the IPCC’s Fifth Assessment Report (IPCC, 2013) includes an update of the understanding and observational evidence of changes in climate extremes. This report has three main parts. The first section reviews the latest findings in climate change science and its effect on extreme weather events. The second outlines what catastrophe modelling is and how it came to be developed. The third section examines whether and how catastrophes models account for climate change through a series of case studies provided by a range of model providers, including AIR, RMS and EQECAT. The appendices provide detail on the terminology used to describe levels of confidence and likelihood (Appendix 1) and the limitations of climate models (Appendix 2). 1 See Appendix 1 for further detail. 5
3 The science of Climate Change The Summary for Policymakers of the IPCC’s Fifth Assessment Report (2013) reports an unequivocal warming of the climate system. Changes are observed in atmospheric and oceanic temperatures, the extent of ice and snow coverage, and the concentration of greenhouse gases in the atmosphere. Many of these changes have been unprecedented over time scales from decades to millennia. Global average air temperatures during the last three decades have been the warmest since 1850 and, in the northern hemisphere, the past 30 years were likely the warmest period for at least 1,400 years. These long term changes are generating widespread impacts, notably: • 2 Increasing accumulation of energy in the world’s oceans: it is virtually certain that the top 700 m of the oceans warmed over the last four decades. • From 1901-2010 global mean sea levels rose by approximately 19 cm. The rate of sea level change th since the middle of the 19 century is larger than the average rate of change over the past two millennia. • There are changes in the mass losses in the Greenland and Antarctic ice sheets, a decrease in the size of glaciers all over the world and a shrinking extent of Arctic sea ice in the Northern Hemisphere. • Atmospheric levels of the greenhouse gases carbon dioxide (CO2), methane and nitrous oxide are higher than at any time during the last 800,000 years. The main causes for this are the combustion of fossil fuels and changes in land use. Since pre-industrial times atmospheric CO2 concentrations have increased by 40% and the world’s oceans have absorbed about 30% of the emitted carbon. This increased uptake by the oceans results in their increased acidification levels. Increased greenhouse gas concentrations, observed warming trends and scientific understanding of the climate system point to a clear human influence on the climate system. Continued emissions of greenhouse gases will result in further warming and are likely lead to changes in many climate system components. 3.1 Temperature extremes The current understanding, which is based on a large body of evidence, indicates that most of the global land areas that were analysed have undergone a significant warming in both minimum and maximum th ii temperature extremes since the early 20 century . An investigation of multiple data sets has shown with high confidence a stronger increase in minimum temperatures than in maximum temperatures on a global scale, and a global decrease in the number of cold nights and days with a simultaneous increase in warm days and nights is very likely. There is however only medium confidence in the reduction of the daily iii temperature range, and the overall impact on probability distributions remains an open question . In contrast to the observed large-scale warming, some regions exhibit changes that are indicative of episodes of local cooling. These regions include central North America, the eastern United States and parts of South America. The difference in the trend for these regions appears to be linked with maximum temperatures that are connected with changes in the water cycle and land-atmosphere interactions and long- term (multi-decadal) variability in the Atlantic and Pacific Oceans. There is only medium confidence that the length and frequency of warm spells or heat waves have globally increased since the 1950s, which is partly due to regionally insufficient data and remaining inconsistencies in the definition of extreme temperature events. Nevertheless, it is considered likely that during this time period the frequency of heat waves has iv increased in Europe, Asia and Australia . 2 See Appendix 1 for common phrases used in the IPCC reports. 6
Catastrophe Modelling and Climate Change 3.2 Precipitation and droughts A generally wetter climate is reflected consistently in changes to precipitation extremes. Recent findings continue to support the earlier assessments that more regions show a statistically significant increase in the number of heavy precipitation events than a decrease. However, the level of statistical significance is lower for precipitation extremes than for extreme temperatures. This is due to spatial patterns of change being less coherent when compared with temperature trends, as well as large areas showing opposite signs in their respective trend. There is strong regional and sub-regional variation in precipitation extremes since 1950. Additionally, it remains difficult to provide a universally valid definition for extreme precipitation events. Only North and Central America and Europe exhibit likely (or higher confidence) increases in either frequency or intensity of heavy precipitation. In Europe and the Mediterranean there remains significant seasonal and regional variability with a large part of the increase occurring during winter (see e.g. Table 2.13 in IPCC, 2013). In Asia and Oceania, regions showing increasing extreme precipitation outweigh those exhibiting a decrease, whereas for Africa a significant trend of extreme precipitation could not be established. In addition, the trends for small scale severe local weather phenomena (such as hail or thunderstorms) are uncertain v due to inhomogeneous historical data and insufficient density of monitoring stations . Assessment of changes in the magnitude or frequency of floods remains difficult. Working Group II of the IPCC’s Fourth Assessment Report stated that no general trend existed in the incidence of floods. River management is an important factor influencing trends in flooding. The strongest indication for flood trends has so far been found in high latitudes of the Northern Hemisphere, although regional variability is such that no clear evidence is currently available. SREX reports an earlier onset of spring flow in regions dominated by snow, however both SREX and IPCC (2013) did not find global trends for flood occurrences, citing lack of evidence. The occurrence of droughts, on the other hand, was comprehensively assessed in the SREX report. SREX concluded that distinction between different types of drought and complex issues with defining droughts have a significant impact on conclusions on global scale trends, and reported with medium confidence that since th the mid-20 century some regions of the world had experienced more intensive and longer droughts (IPCC, 2012). Due to the scarcity of direct measurements of soil moisture and other variables related to drought often other related variables hydrological proxies are used for drought assessments. The severity of an assessed drought event is highly dependent on the choice of variable and the length of time scale considered. Agreement is however found for some regions. There is high confidence of increasing drought in the Mediterranean and West Africa and also high confidence is reported for a decrease in drought for central North America and north-west Australia. 3.3 Tropical and extra-tropical storms Tropical and extra-tropical storms account for the highest impact extreme events. There is limited evidence for a long-term trend in the number of tropical storms globally. In addition to frequency or storm count it is necessary to consider the associated changes in the intensity and duration of tropical cyclones. The quality of observations has changed significantly over the past century, for instance after the availability of satellite data. Measurements of storm intensity are very sensitive to observation technology and therefore long-term historical trends are influenced by changes therein. Regionally it is virtually certain that since the 1970s there has been an increase in the frequency and intensity of the strongest tropical cyclones in the North Atlantic vi basin . However over the last 100 years there have been other periods of high activity. The variability of trends makes a confident attribution to climate change challenging, although there are good physical reasons to expect hurricanes to be stronger on average. There is limited evidence for a change in extra-tropical storms or extreme winds globally. Long-term wind measurements exist often for too short periods (particularly in the Southern Hemisphere) or are inconsistent due to changes in observation technology in order to derive long-term records. Therefore proxy data is commonly used such as in-situ surface pressure observations or pressure data from reanalyses to derive changes in the geostrophic wind field. In the latter case the findings are sensitive to the reanalysis product with newer generation products typically outperforming earlier reanalyses. Studies using reanalysis data suggest a northward and eastward shift of Atlantic cyclone activity, with intensification of cyclones during vii winter and at high latitudes . 7
4 Catastrophe modelling 4.1 Evolution and Role of Catastrophe Modelling in the Insurance Industry Catastrophe modelling is a comparatively young discipline, with its origins embedded within property insurance and the science of natural hazards. It aims to help companies anticipate the likelihood and severity of potential future catastrophes before they occur so that they can adequately prepare for their financial viii impact . Commercially available catastrophe models (often referred to as ‘cat models’) have only been in existence for the last 25 years. Prior to the birth of catastrophe models in the late 1980s, rudimentary methods were adopted for estimating catastrophe losses. Standard actuarial techniques were not appropriate to estimate future catastrophe losses. Historical loss data was, and still is, scarce, especially for low frequency high severity events with the potential to threaten insurer solvency. For risk acceptance, underwriters used spatial risk mapping and measuring of hazards, but these had traditionally been performed quite separately. For pricing, they relied either upon internally generated Probable Maximum Losses (PMLs) using rule-of-thumb formulae, or upon realistic estimations of potential loss using subjectively derived deterministic scenarios. There was a clear focus on the severity of potential events, but no reference to frequency. At this time the unfulfilled aspiration for simultaneous consideration of these elements was recognised by those responsible for founding three of the globally recognised catastrophe modelling software providers: AIR (1987), RMS (1988) and EQECAT (1994). Despite the commercial availability of the first probabilistic catastrophe models in the late 1980s, their use was not widespread. Reinsurance cover was readily available and the market was somewhat benign. Meanwhile the software providers were generating large probabilistic US hurricane industry loss estimates of ix USD $20-30bn occurring with a reasonably significant probability . In 1989, the magnitude of loss caused by x xi both Hurricane Hugo ($4bn ) and the Loma Prieta earthquake ($6bn ) sparked initial interest in the use of catastrophe models amongst insurers and reinsurers. However, it was the unprecedented and unforeseen loss size from Hurricane Andrew in 1992 which really highlighted deficiencies in the purely actuarial approach to quantifying catastrophe risk losses. In real-time, AIR issued a fax to its clients estimating losses in excess of $13bn based upon the AIR hurricane model. Months later, the Property Claims Service (PCS) xii reported an industry loss of $15.5bn . Losses of this size hit the market hard, resulting in the insolvency of xiii 11 insurers . As a response, adoption of catastrophe models grew exponentially as they were viewed as a more sophisticated and reliable approach to catastrophe risk assessment. Increasing population densities and property values in hazard prone areas has led to diversification in the use and coverage of catastrophe models. Catastrophe modelling technology is now used extensively by insurers, reinsurers, governments, capital markets and other financial entities. Catastrophe models are an xiv integral part of any organisation that deals with natural catastrophe risk , and are used most commonly to perform activities such as risk selection and underwriting, reserving and ratemaking, development of mitigation strategies, design of risk transfer mechanisms, exposure and aggregate management, portfolio optimisation, pricing, reinsurance decision-making and capital setting. Catastrophe models are developed by harnessing loss and hazard observations, building upon existing data, testing existing models and incorporating these lessons into future catastrophe modelling advances. Recent developments include explicit modelling of windstorm clustering, better understanding of inter-country demand surge relationships, potential for occurrence and impact of storm surge, response of business interruption losses, development and damage of hurricanes beyond coastal areas and appreciation of non- modelled components of catastrophe models. The events of 9/11 also drove the first development of models for man-made catastrophes, in the form of terrorism models. By their nature, models represent an approximation of expected outcomes, and they are just one of many tools used to enhance the understanding and management of risk. Newly available loss data, increasing understanding of the science of natural hazards, and advancements in computing capability and technology all contribute to the evolution of catastrophe models and the dynamic discipline of catastrophe modelling. 8
Catastrophe Modelling and Climate Change 4.2 How does a catastrophe model work? Catastrophe modelling software contains a vendor-specific view of hazard risk, and vulnerability of insured property. This view is devised using observed data as a basis. The software then facilitates application of this view of risk to a specific client’s book of business in order to quantify probability and size of potential loss. This is achieved by reducing the complexity inherent in physical interaction between hazard and vulnerability, by parameterising characteristics to a limited set of measurable metrics. These metrics are applied systematically, consistently and repeatedly to a custom set of exposure data. The insurance related financial characteristics can then be overlaid to give a net loss bespoke to the client using the tool. Most catastrophe models accomplish this by adopting a modular approach (Figure 1). Figure 1: Adapted from Dlugolecki et al., 2009 Operating the software is however only a small part of what it takes to optimise the use of catastrophe modelling within a business. It is imperative that those responsible for running the model can also effectively understand, interpret and convey the output with consideration for the models’ limitations. Catastrophe models can provide a variety of financial outputs, the most common of which are the Annual Average Loss (AAL) and the Exceedance Probability (EP) Curve. The AAL is sometimes also referred to as the ‘pure premium’ or ‘burning cost’ and can be incorporated into pricing along with an allowance for expenses and recommended return on capital. The EP Curve is commonly depicted as a graphical representation of the probability that a certain level of loss will be exceeded. Reading off points on the curve will give different views on frequency and severity of losses for the book of business being modelled. These curves are valuable for insurers and reinsurers to determine the size and distribution of their portfolios’ xv potential losses . Even though catastrophe models are sophisticated, they cannot capture the full spectrum of risks that exist in the real world. Each catastrophe modelling vendor will have a suite of models covering region/perils that are of interest to their clients. No model is currently capable of covering every peril in every region, leading gaps in model availability which can be termed ‘model miss’.. Where a catastrophe model does quantify losses for a region/peril, the process is complex and depends on many assumptions which naturally result in a degree of uncertainty around that loss. This uncertainty increases for more extreme events where there is little empirical experience and instances where exposure data imported into the catastrophe model by the client is of poor quality. It is paramount that the limitations of the model and the uncertainty inherent in its outputs are conveyed effectively during the decision making process. In order for catastrophe models to assist in the forecasting of risk exposure, they must incorporate observed trends. The next section provides a series of case studies to examine how the catastrophe modelling community is addressing the challenge of long term climate change trends. 9
5 Catastrophe models and Climate change Having reviewed the latest findings on climate change and outlined the workings behind catastrophe modelling, this section details a series of studies from catastrophe model providers across a range of natural hazards in order to assess the extent to which climate change influences the outputs of catastrophe models. In the case study provided by EQECAT it is shown that a northward shift of European storm track and intensification of the strongest storms is projected by at least one global climate model. Madeleine-Sophie Déroche of the Climate – Knowledge Innovation Centre points out that the intensification is a consistent feature amongst the latest generation models, but systematic track changes are shown to vary depending on the model and region being analysed. UK flood modelling by JBA supports research by the UK Government which shows that climate change could almost double the number of properties at significant risk of flooding by 2035 unless there is additional action. It is argued by several case studies that any recent climate trends will implicitly be included in the data that is used to construct catastrophe models. Three case studies address cyclone hazard. An attempt at medium–term (1-5 years) forecasting is shown by RMS which includes sea surface temperature trends as a predictor for anomalous Atlantic Hurricane activity. They find a small adjustment of the number of cyclones using this method. A more significant effect of climate change is found when considering that as much of 30% of the surge contribution to losses from Superstorm Sandy can be attributable to long-term changes in sea-level. AIR also found modest increases for cyclone losses in the South Pacific, but found no compelling evidence for a climate change signal in US thunderstorms. James Elsner from Climatek makes the case for increased losses of 5% per decade in the North Atlantic driven by trends in sea-surface temperature. He also points out that climate models underestimate the sensitivity of cyclones to temperature. The three independent tropical case studies and approaches show consistency in the direction and order of magnitude of the changes projected. Note that the case studies reflect the views and opinions of the contributors – in any one case the views don’t necessarily reflect the views of Lloyd’s or the other contributors. 10
Catastrophe Modelling and Climate Change European Windstorm By Iain Willis, Product Manager at EQECAT Origins Every winter, deep low pressure systems originating in the mid and high latitudes of the Atlantic give rise to intense Extra Tropical Cyclones (ETC). Unlike Tropical Cyclones (such as Typhoons or Hurricanes), European windstorms usually begin along a polar front where a body of low pressure from the North encounters high pressure from the South. The movement of these opposing air masses creates a cyclonic shear, which, if given the right circumstances, can cause an anti-clockwise rotation of the air mass around a core of low pressure. Following this process of ‘cyclogenesis’ (Figure 2), the Extra Tropical Cyclone tracks eastwards, driven by the fast moving Northern Hemisphere jet streams (and at speeds ranging from between 30-70mph). Since Extra Tropical Cyclones are frontal systems, largely driven by the temperature and pressure contrasts of the mid-latitudes, European windstorm frequency and severity is typically greater during the winter months. Figure 2: The development of an Extra-tropical Cyclone (ETC) Studying the path of European windstorms using sources such as remote sensing data, historical wind speeds, and subsequent reanalysis (e.g. ERA-Interim), the data shows that the most common trend is for storms to track across the Northern latitudes of Europe, primarily impacting the UK, Ireland and Scandinavian countries. Such a trend might suggest that countries in the lower latitudes of Europe are not exposed to windstorms. This is not the case. European windstorms can deviate from this norm and often with xvi devastating consequences. As seen in Figure 3, highlighting the storm tracks of multiple industry loss events, windstorm Xynthia (February 2010) emanated from the lower latitudes of the Atlantic, before tracking xvii North-Eastwards across Portugal, Spain and France, causing a total of 2.9bn USD in insured loss. 11
Figure 3: Storm tracks of major industry loss events and European population density (Source: EQECAT / Storm tracks based on the Extreme Wind Storms Catalogue (XWS), University of Reading, 2013) Unlike Hurricanes, which tend to have a defined, central core of low pressure, European windstorms are often more disparate in structure. Similarly, a European windstorm’s size and structure can greatly affect its damage potential and footprint. For example, windstorm Daria (January 1990) had a very large footprint of damage whereas windstorm Klaus (January 2009) rapidly intensified as it tracked across continental Europe, creating a small but very damaging radius of strong winds. Frequency and industry loss Between 1959-2001 reanalysis data shows an average of about 19 windstorms a year over Europe. Of these events however, we would only expect around a third of these storms to result in an insured loss. The most severe and frequent occur primarily between the months of December and February although the impact of an early season event, such as Windstorm Christian (October 2013) or the Great Storm of ‘87 (October 1987), can be more damaging as most trees still retain their leaves at this time of year. Within the historical data there is considerable variability both in the number of storms and their relative severity. The winter of 12
Catastrophe Modelling and Climate Change 1989/1990 had the highest windstorm frequency with 37 events recorded. It proved a notable year as several th of the windstorms were significant from an insured loss perspective, such as windstorms Daria (25 th th January), Vivan (26 February) and Wiebke (28 February). Such seasonal intensity is not in isolation and also took place during the 1999 storm season as Anatol, Lothar and Martin caused a total loss of $13.9 xviii billion . Lothar and Martin occurred within 36hrs of each other, both impacting similar areas of central Europe. Such frequency is commonly referred to as ‘temporal clustering’. Due to the magnitude and aggregation of insured losses in such years, the pattern and periodicity of temporal clustering is of considerable interest to the insurance market. As traditional excess of loss reinsurance contracts are often based around the 72 hours clause, the impact of having multiple meteorological events occurring within this period causes major concern as reinsurance trigger levels may be reached by the aggregation of these smaller events, rather than by a single, very large windstorm. European windstorm severity is currently assessed by meteorologists using a Storm Severity Index (SSI). A single value is calculated for a storm event based on several physical characteristics of the event. This typically includes factors such as the maximum observed wind speed, a minimum wind speed threshold (e.g. 25 m/s), storm duration, and the physical extent of the storm over land areas. By using such metrics, meteorologists are able to assess the relative severity of these large complex cyclones. It is important to note that there remains no universal standard for calculating SSIs and they continue to be a moot point for meteorologists. Studying the insured losses from single historic windstorms, the most significant events were recorded from xix Daria (1990) and Lothar (1999) with insured losses totalling $8.2 billion and $8 billion respectively th However, the most recent large scale loss from a windstorm was in 2007 when Kyrill (18 January) impacted xx much of central Europe causing $6.7 billion insured loss . The common factor in the characteristic of the most damaging windstorms concerns their international impact, i.e. a single storm can impact multiple countries as it tracks across central Europe. Due to the very high wind speeds of these storms (typically found to the south of the storm vortex) and the large synoptic size (~1000km), destruction to property, contents, and business downtime is on a very large scale. As the storm track of Kyrill shows (Figure 3), it was able to cause considerable damage across multiple European countries including Austria, Belgium, France, Germany, Ireland, Netherlands, Poland, the Czech Republic and the United Kingdom. Climate modelling In considering how climate change will affect the pattern of European windstorms, it is necessary to make use of Global Climate Models (GCM). These state-of-the-art numerical weather models have become an integral tool in meteorological research and help scientists simulate future climates in line with IPCC 3 emission scenarios (SRES) . EQECAT have worked closely with the Free University of Berlin in using the ECHAM5 coupled climate model TM as well as our European windstorm catastrophe model (Eurowind ) to study the meteorological and financial impacts that future climate conditions may bring. Comparing twenty-five different storm parameter th statistics (including storm severity, wind speed, area, duration, and atmospheric pressure) of the entire 20 st Century with the 2007 SRES emission scenarios (A2 and A1B) for the 21 Century, the use of a GCM has provided us with valuable insight into the possible impacts of climate change on European windstorm track density and severity. Results were calculated at five different meridional European transects spread evenly o o across the longitudes of 0 E to 20 E. It is important to note that ECHAM5 is one several GCMs used by the IPCC community and therefore several scenario runs cannot capture the full spectrum of uncertainties associated with climate change simulations. ECHAM5 results have been shown to lie in the middle of all ENSEMBLES simulations, but the results presented here must still be seen as one of many possible climate change outcomes. Despite the differences in various GCM research methodologies, model assumptions, resolution and the magnitude of storm behaviour, there has proved to be consensus between EQECAT’s results and other xxi,xxii, xxiii,xxiv published research in this field. Other climate change findings have all noted that changes in near-surface temperature, baroclinicity (low level temperature contrast) and sea-ice will greatly affect the strength and location of mid-latitude cyclone activity. On a global perspective, such changes may result in a 3 Control run was based on CO2 emissions from 1900-2001. 13
gradual polar shift in storm track density. With regard to windstorm activity in the European mid-latitudes, EQECAT’s research highlighted the following observations in future climate scenarios: 1. An increasing volatility - fewer smaller storms but an increase in the frequency of very large storms; 2. A shift in the latitude of European windstorms towards central Europe (between bands 48N-61N); and 3. A four-fold increase in the frequency of years with several severe storms (this is based on the normalised equivalent of an SSI sum of three times Daria sized storm events per year). Figure 4: Possible climate change impacts on European windstorm (Clockwise from Top Left: a) Increasing severe storms b) Correlations between SSI and storm parameters c) Frequency of years with equivalent SSI of 3 4 x Daria-sized storms d) Location of storm activity) In considering the consequences of these changes to insured loss, the overall impact on European exposure from these results implies a 3-5% decrease in the total number of potentially damaging storms but a 10-20% 5 increase in the number of larger storms . Likewise, the progressive shift of storm tracks to the central latitudes in Europe could increase the severe storm losses seen in major European markets, disproportionately impacting France and Germany. 4 Findings are based on EQECAT whitepaper ‘Activity of Catastrophic Windstorm Events in Europe in the 21st Century’ (2011). 5 Defined as having an SSI equal to or greater than windstorm Daria. 14
Catastrophe Modelling and Climate Change Modelling Various techniques are employed in modelling ETC. These range from solutions such as pressure field analysis, using historically recorded wind speed data, to the use of supercomputers and Numerical Weather Prediction (NWP). In recent years, and given the rapid advancement of IT, meteorologists and modellers are increasingly turning their attention to NWP and GCMs. NWP can essentially be split into macro and meso-scale models. Although both are computationally very intensive and based on the same mathematical equations of thermo dynamics, the subsequent geographic scale, complexity and end-purpose of these models varies considerably. Meso-scale NWP modelling is generally at a very high resolution but is only run for short periods of time. For this reason it is typically used for short-term weather forecasting. A good example in the progression of this science was evidenced in October 2013 by the European forecasts that predicted with considerable accuracy the track and intensity of windstorm Christian (also referred to as the St Jude’s day storm in the UK) several days before it made landfall. Macro scale modelling concerns the use of GCM and Atmosphere-Ocean General Circulation Models (AOGCM). These models can be used to simulate weather on a global scale for thousands of years. Unlike meso-scale techniques, AOGCMs take into account the changing condition of global climate controls such as the extent of sea-ice, vegetation coverage, and the complex interaction between the oceans and the atmosphere. In doing so, they are extremely useful for studying longer-term processes such as natural and anthropogenic climate change, as well as key weather pattern signals (e.g. NAO, AMO, ENSO). European windstorm modelling has evolved considerably in recent years. Given that historically observed wind speed data has only been recorded for the last 50 years, and with varying granularity, it provides only a small window into the past with which to understand the frequency of this complex peril. However, given the 6 advances in GCMs, probabilistic analysis, the reanalysis of historic data, and the downscaling techniques EQECAT is now able to employ, European windstorm modelling is rapidly evolving. TM In creating Eurowind (a fully probabilistic risk model that quantifies prospective risk from windstorms in 24 countries across Europe), EQECAT makes use of these latest capabilities. We combine over 50 years of historically observed wind speed data (from thousands of meteorological stations across Europe) with a 1200-year climate simulation from using an AOGCM to inform on the key parameters of this peril. Historic storm seeds are used to develop a stochastic catalogue of synthetic storms. Using a perturbation approach, theoretical distributions are fitted to major parameters of historical storms (e.g. SSI, duration, track, severity) to probabilistically generate new storms. This historic and synthetic storm catalogue features ~20,000 events. Making use of a state-of-the-art AOGCM model, we are able to validate key metrics, such as storm frequency, clustering, and the spatial coverage of European windstorms. Likewise, in downscaling these large cyclonic events, we employ a combination of both deterministic and probabilistic modelling techniques. For example, the latest global land use information and digital elevation data (DEM) are combined to accurately modify wind behaviour, together with local gust parameterization to model the damaging surface winds of an Extra Tropical Cyclone. EQECAT does not incorporate future climate change scenarios into its standard catastrophe models. It would be largely premature to do this given the enormous volatility in catastrophic event activity on annual and decadal timescales. However, given the use of historically recorded wind speed data over the last 50 years in constructing EQECAT’s European windstorm model, climate variability during this time could be considered to be implicitly built into the model. From an industry perspective, EQECAT’s paradigm is to remain focused on our clients' need to write business in today’s market and concentrate on quantifying the current risk via the best available science and data. 6 Downscaling methods are developed to obtain local-scale surface weather from regional-scale atmospheric variables that are provided by GCMs. 15
European Windstorms By Madeleine-Sophie Déroche, Junior Analyst at Climate-KIC / LMD / LSCE European winter windstorms associated with extra-tropical cyclones ETCs are one of the main atmospheric phenomena of the mid-latitude regions, where they are responsible for episodes of strong surface wind speeds and rainfall. In the northern hemisphere, the development of such systems is favoured during the autumn and winter seasons (October to March), when the temperature difference between the Equator and the Poles is the strongest. Intense ETCs associated with extreme surface wind speeds (also called windstorms), are generated over the North Atlantic region and sometimes reach Western and Central Europe. They can cause wind-related damage as well as flood events (Kyrill, 2007) and storm surges (Xynthia, 2010). The dynamical processes at the origin of the development of extreme ETCs have been studied either by xxv,xxvi, xxvii focusing on one specific windstorm that caused important damages or by comparing the xxviii,xxix meteorological situation for a group of extreme ETCs . A key process that emerges is the interaction and mutual amplification of anomalies at the upper and lower levels of the troposphere through vertical motions. Extreme ETCs are more frequent and more intense during the positive phase of the North Atlantic Oscillation (i.e. main atmospheric pattern driving the weather over Europe in winter), and are more likely to reach Europe with a strong polar jet stream (i.e. eastward wind at 11 km height). Observed trends in European winter windstorms in the recent past The assessment of windstorm trends depends on the type of data that are used. Studies that consider storminess over the North Atlantic region (either from wind speed or pressure measurements from meteorological stations) find significant variations at decadal and longer time scales, with a minimum of storminess around 1960 and a maximum around 1990. Values of storminess at the beginning of the 21st xxx,xxxi, xxxii century are as high as those at the beginning of the 20th century . However, when looking at trends in specific countries and for shorter time periods, one finds that local variations do not coincide with the ones xxxiii, xxxiv from the more extended North Atlantic region . This highlights the importance of the chosen geographical area and the length of available records when dealing with trends of atmospheric processes with a high natural variability. Reanalysis datasets, in other words assimilation of observations into climate model, are another type of meteorological data that are used to analyse the recent past. Studies focusing on xxxv, xxxvi storminess over Europe find an increase over different periods of time . Observed increasing trends in wind-related damage during the last decades seem to be mainly due to an increased vulnerability of the population and of natural environments. On the one hand, the increase in economic losses associated with windstorms can be explained by the growth of insured population in xxxvii, xxxviii exposed areas . On the other hand, a warming climate may impact natural environments, making them more vulnerable to windstorms. In Switzerland, warmer and wetter winters have been observed over xxxix the last decades impacting the quality of soils and favouring forest damages during windstorm events . Global Climate Models GCMs compute the three-dimensional time evolution of the atmosphere and the ocean, including the wind or current, temperature and moisture content. The models divide the atmosphere into “boxes”, where each meteorological variable is represented by one value. The equations of motion are then discretized over these boxes and solved; processes that occur at a scale smaller than the box size are represented implicitly. The performance of GCMs follows the technological evolution of supercomputers, leading in particular to more and smaller boxes. An advantage of using GCMs for weather-related risks is the opportunity to analyse all the factors that xl contribute to or hinder the development of a specific phenomenon . The low resolution (large boxes) of previous generations of GCMs led to a poor simulation of ETCs, but many current GCM simulations now yield a sufficient resolution to investigate ETCs from both physical and impact perspectives. The comparison xli with reanalysis datasets for the current period indeed shows that ETCs are now well simulated . Storm tracks over the North Atlantic region are well represented even though there are still some biases affecting xlii the number of ETCs reaching Europe and their intensity . 16
Catastrophe Modelling and Climate Change Climate Change modelling: from SRES to RCPs Since the 1990s, the IPCC has coordinated a worldwide assessment exercise consisting of a review of published scientific research papers. Synchronised with the IPCC reports, a number of model inter- comparison projects (MIP) use GCMs from major worldwide institutions to assess the impact on climate of both past and future emissions of greenhouse gases (GHG) and aerosols that change the energy balance of the Earth. The modelling of future climate change and its impacts is based on projected emission scenarios. Since the first IPCC assessment report (FAR) in 1990, four sets of emissions scenarios have been released xliii and the approach used to define them has evolved . The three first sets were based on assumptions on the potential evolution of economic and population growth, energy supply and consumption patterns, development of clean technologies and undertaken climate policies. In 2000, the scenarios from the Special Report on Emission Scenarios (SRES) encompassed a wider range of uncertainties and took into account xliv,xlv,xlvi more social, political and economic factors than scenarios back in 1990 . From these socio-economic storylines, time series of GHG emissions were derived and used to force GCMs. It took several years to complete this sequential process, leading to a high risk that the basic socio-economic assumptions would change, be verified or proven false. Since 2008, a new approach to design emission scenarios has been chosen for the purpose of the last IPCC xlvii assessment report (AR5) . The new emission scenarios, named Representative Concentration Pathways (RCPs), are now based on the identification in the existing literature of possible levels of radiative forcing by 2100. Socio-economic assumptions do not underlie the RCPs and possible paths leading to these levels of forcing are thus no longer restricted to only one per emission scenario. Projected trends from models participating to the fifth Coupled Model Intercomparison Project (CMIP5) The latest model inter-comparison project (CMIP5) a worldwide coordination between climate institutes running their GCMs in the purpose of the assessment of climate change impact. A study comparing the results from different ETC detection methodologies applied to one CMIP5 model finds an increase in the xlviii number of intense ETCs in the Northern Hemisphere . However looking at the response of the storm tracks to climate change, results differ from one model to the other. In order to explain these differences, it is xlix possible to analyse in detail within each GCM the processes surrounding the occurrence of ETCs . It has been shown that changes in the meridional (i.e. northward) and vertical gradients of the temperature have competing effects and it is difficult to assess what the overall impact would be. Another important factor is the increase in atmospheric moisture content that provides a latent heat reservoir increasing the potential intensification of individual ETCs, but also will have a general weakening effect on the development of ETCs by enhancing the energy transport from the Equator to the Pole and reducing the temperature difference. 17
UK Flood By Professor Rob Lamb, Chief Scientist, Richard Wylde, Meteorologist, and Jessica Skeggs, Hazard Mapping Specialist at JBA Group The evidence for change in the UK Although winter rainfall totals have changed little in the last 50 years and annual totals appear not to have changed significantly since records began in 1766, over the last half century an increasing proportion of the l UK’s winter rain has fallen during intense wet spells. Additionally, a report on historical trends in the UK climate concluded in 2008 that sea levels have risen around the UK over the past century. Past and present greenhouse gas emissions mean further climate change is inevitable in the next 20 to 30 li years no matter how emissions change in the future. Current UK climate projections are based on the analysis of outputs from multiple climate models and scenarios for future greenhouse gas emissions. However, predicting the influence of human-induced change is complicated by the internal variability in the climate system. Over the next few decades, this internal variability is a major source of uncertainty. It is estimated to account for nearly half of quantifiable uncertainty associated with regional winter precipitation lii changes for the period 2010-2019, compared with only a quarter of the prediction uncertainty by the 2080s . Even so, statistically significant changes in accumulations of winter rainfall (the type of weather pattern liii associated with some major recent floods such as in the winter of 2000) could be detectable by the 2020s . For the shorter intense storms that are associated with surface water flooding, there is more variability and so it will take longer to detect significant trends. Despite this, there are accepted physical mechanisms by which an increase in extreme rainfall should be expected and emerging observational evidence in support of liv this . Long term records of river flooding also paint a complex picture. River flows are affected by changes such as land use and drainage works as well as by the climate, and there is little statistical evidence for long-term lv trends in maximum river flows over the last 80 to 120 years despite many shorter-term fluctuations . This evidence does not rule out the possibility that climate change has affected river flooding, or that it will do so in future, although it does illustrate that long-term trends are difficult to identify when set against the variability of river flow records. lvi Recent studies have used climate models to attribute specific flooding events, at least in part, to climate lvii change. For the UK it has been suggested that there was very likely to have been at least a 20% increase in the risk of occurrence of the winter of 2000 floods (which damaged almost 10,000 properties and led to insured losses worth an estimated £1.3 billion, or £1.85 billion at today’s values) associated with human- induced climate change. These conclusions are dependent on a complex chain of models and assumptions, and may not necessarily reflect future changes. However, they do provide emerging evidence of an increasing risk of inland flooding in the UK and form part of a growing international evidence base on the lviii attribution of extremes – despite individual weather events having not been previously attributed to climate change. Future climate change Over longer time horizons, the greenhouse gas emissions are expected to have an increasing influence on predicted climate in the UK. By the 2050s, there is clear divergence in the latest UK climate change projections between alternative future greenhouse gas scenarios. By the 2080s, UK projections shows that there could be around three times as many days in winter with heavy rainfall (defined as more than 25mm in lix a day) as in the recent past. It is plausible that the amount of rain in extreme rain storms (defined as storms with a 1 in 5 annual probability, or rarer) could increase locally by 40%. The impact of climate change on river flooding is highly dependent on the geography of the river catchment. Research based on climate models that have been used in the UK projections shows a range of increases in peak river flows from 10% to 15% over the period between 2015 and 2039, rising to a range of 20% to 30% lx by the 2080s . Flood risk is, of course, also affected by property development and by investment in flood risk management measures. Sea levels around the UK are projected to rise too. The predicted changes depend on what assumptions are made about future greenhouse gas emissions, and on the method used to characterize scientific uncertainty 18
Catastrophe Modelling and Climate Change in the modelling. For the UK climate projections, central estimates of sea level rise for a medium emissions scenario (a world in which there is less, but still increasing reliance on fossil fuels) are in the range 24-36 cm lxi by the 2080s, relative to 1990 levels . However under a high emissions scenario (in which reliance on fossil fuels continues to increase rapidly) there is a chance that sea level could rise by approximately 70 cm. There are credible (though unlikely) extreme scenarios that would lead to even greater increases in sea level of up to 1.9m. What does this mean for flood risk? The UK Government’s Adaptation Sub Committee has reported that “current levels of investment in flood defences and in the uptake of protection measures for individual properties will not keep pace with the increasing risks of flooding. Climate change could almost double the number of properties at significant risk lxii of flooding by 2035 unless there is additional action.” The Environment Agency estimates that investment needs to increase by £20 million above inflation every year merely to stand still and avoid increasing the risk associated with climate change and deterioration of flood defences. Increased investment to counter projected climate change could bring about a four-fold reduction in the risk of flooding compared with a scenario of no additional action. Do flood risk models account for climate change? The UK climate projections are gradually being incorporated into flood risk models used for the purposes of local flood risk management and investment plans. Assessments of risk at national scale are also building in lxiii climate change to inform long-term investment strategy . The UKCP09 projections provided a step change in the available detail, and for the first time a probabilistic treatment of the scientific uncertainty in the climate models. There are case studies demonstrating the application of the probabilistic projections in assessing lxiv,lxv future flood risk , but these require advanced modelling methods and are not yet routinely applied. Box 1 demonstrates the sensitivity of flood risk to future climate change for a part of the River Thames basin, based on detailed flood risk mapping and scenarios derived from the latest UK climate projections. In the insurance industry, probabilistic catastrophe models from companies such as AIR, EQECAT, JBA and RMS are available to assist insurance companies assess their likely losses due to flooding in the UK. Most catastrophe models are based on long term data which may have a signal of climate change from the recent past inherently present within them. However, uncertainties associated with the estimation of both the degree to which climate change is occurring and to the resulting change in local flood severities and frequencies means that the impact can be difficult to account for in risk models. The catastrophe models used today to consider UK flood risk (both river and sea surge) do not, therefore, explicitly model the impact of future climate change. Forward look Future climate change is inherently uncertain because we cannot be sure about how greenhouse gas emissions may be changed by economic development, technology and government policies around the world. However climate models are driving towards greater spatial detail and better resolution of key physical processes. Over the next decade it is possible that weather forecasting and medium-term climate models may start to converge, giving better foresight of weather at the annual scale and more precise projections about how the climate will respond over the longer term to changes in the atmosphere caused by human activity. Allied to these changes, we can expect progress in the representation of internal variability, and a view of future climate that includes better knowledge of variability as well as of average conditions. There should be more explicit accounting for uncertainties, especially in predictions made by hybrid models that integrate climate, weather and flood risk more closely than at present. Models used to predict flooding will become more efficient and able to provide better information at property level. Side-by-side with these developments, JBA can anticipate greater integration of projections of climate with other changes (such as lxvi, lxvii land use change, flood mitigation investments and industry losses). Studies have already stressed the need to quantify climate change in catastrophe modelling. It is anticipated that methods to do this will develop if, as expected, the effects of climate changes become more apparent over the coming decades. This process must be properly structured and make good use of the evolving climate science, and so communication between researchers, catastrophe modellers and insurers is expected to be important. 19
You can also read